SYNOPSIS
abxtest [-n min] [-m max] [-g goal] A-cmd B-cmd
DESCRIPTION
abxtest is a tool for conducting listening (or other sub
jective) tests to determine whether a listener can discern
a difference between two subjects under test, denoted A
and B.
The listener conducts several trials in which an unknown
subject, X, is randomly chosen to be either A or B. The
probability that X is A or B is the same. The listener
must decide whether X sounds the same as A or the same as
B. The listener may repeat A, B, and X as many times as
necessary before reaching a verdict.
In order to prove the hypothesis that a difference can be
discerned between A and B, the listener must disprove the
null hypothesis that A sounds the same as B. This is
accomplished by correctly identifying X often enough that
randomly guessing correctly the same number of times would
be statistically improbable.
The A and B subjects for abxtest are actually arbitrary
shell commands. The normal mode of use would be to give
commands which play two slightly different sounds. The
standard output and standard error from each command are
suppressed so as not to reveal the identity of X. The
listener may interrupt an executing command (usually by
pressing ^C) without terminating the test.
OPTIONS
-n min Set the minimum number of trials. Trials will con
tinue after this number if the statistical proba
bility goal has not yet been met. The default is
10.
-m max Set the maximum number of trials. Trials will end
after this number even if the statistical probabil
ity goal has not been met. The default is 20.
-g goal
Set the minimum statistical probability goal. This
is the probability that the number of correct iden
tifications is the same as random guesses. In order
to disprove the null hypothesis, the result of the
test must be a probability less than or equal to
this goal. The default is 0.05.
FILES
/dev/random
By default this file is read to generate a random
Robert Leslie <rob@mars.org>
MAD 16 November 2000 ABXTEST(1)
Man(1) output converted with
man2html